Skip to main content

Examples

Complete, runnable workflows for real jobs, in Python and TypeScript. Each page is self-contained and uses the real SDK surface end to end (Pareta.from_env() / Pareta.fromEnv()): deploy a model (no GPU knob), call it with OpenAI-compatible inference, and read the metered cost in dollars off run.cost. Grouped by what you are trying to do.

Deploy and call a model​

You know the task; you want a live endpoint and a response.

  • Deploy a model and call it — the two-call workflow: endpoints.deploy(task, model="recommended", wait=True) then chat.completions.create(model=endpoint.id, ...). Covers deploy events, streaming, metering and InsufficientCreditsError, errors, endpoint ops, and async.

Pick the right model first​

You have a job in plain English, or your own data, and want to deploy the model that actually wins on it.

  • From a sentence to a deployed winner — the full funnel: tasks.match to leaderboard to evals.runs on your own data, pick the best kind == "open" model, endpoints.deploy it, then run inference.
  • Benchmark models on your own data — build an eval set from your rows, run open candidates against frontier="benchmarked", and read ranked results plus run.cost.
  • Document extraction (PDF/image) — the blob-task loop: build an eval set from your PDFs/images, upload_document per row, run against open candidates plus vision frontier baselines, pick the winner by quality and cost, deploy, then run OpenAI-compatible inference.

Inference patterns​

Getting tokens out efficiently.

  • Streaming chat completions — stream tokens with chat.completions.create(stream=True): iterate ChatCompletionChunk objects, read delta.content, accumulate full text, plus async streaming and metering behavior.
  • Concurrent calls with AsyncPareta — fire many inference and eval calls concurrently with AsyncPareta and asyncio.gather, with semaphore backpressure and per-task error handling.

Operate and monitor​

Watching what is deployed, and what it costs.

  • Cost & quality monitoring — read what calls and eval runs cost, the open-vs-frontier savings framing, and watch a live endpoint's spend and quality via endpoints.metrics().

Migrating in​

Already on the OpenAI SDK.

  • Migrating from the OpenAI SDK — keep using the openai client against Pareta (base_url + pareta_sk_ key), and when to switch to the pareta SDK for deploy, eval, and discovery.

See also​

  • Concepts and step-by-step explanation: Guide.
  • Field-by-field API docs: Reference.